Recent discussions among researchers highlight ongoing challenges faced by large language models (LLMs) in reasoning tasks. A new paper critiques the effectiveness of current standardized benchmarks, suggesting they fail to accurately reflect the reasoning capabilities of LLMs and expose their weaknesses. One researcher noted that while companies like OpenAI claim high performance on reasoning benchmarks, LLMs struggle with simple tasks, such as the 'Alice' task. Another researcher proposed an innovative method to evaluate LLM reasoning beyond mere accuracy, focusing on positional bias in multiple-choice questions. This approach aims to discern whether LLMs truly understand logic or are merely making educated guesses. Additionally, findings indicate that LLMs could enhance their reasoning by refining their training data through self-generated reasoning paradigms, utilizing a universal text template for training.